EfficientNet B4 RA2
| Property | Value |
|---|---|
| Parameter Count | 19.5M |
| Model Type | Image Classification |
| License | Apache-2.0 |
| Training Data | ImageNet-1k |
| Image Size | Train: 320x320, Test: 384x384 |
| GMACs | 3.1 |
What is efficientnet_b4.ra2_in1k?
EfficientNet B4 RA2 is an advanced image classification model that implements the EfficientNet architecture with RandAugment optimization. It represents a carefully balanced trade-off between model size and performance, featuring 19.5M parameters while maintaining high accuracy on ImageNet-1k classification tasks.
Implementation Details
This model utilizes the RandAugment (RA2) recipe, which evolved from the original EfficientNet augmentation strategies. It employs RMSProp optimization with TensorFlow 1.0 behavior and implements EMA weight averaging. The learning rate follows a step-based exponential decay with warmup period.
- Advanced feature extraction capabilities with multiple output stages
- Optimized for both classification and feature backbone usage
- Supports variable input resolutions with test-time optimization
- Implements modern training techniques from "ResNet Strikes Back" paper
Core Capabilities
- Image Classification with 1000 ImageNet classes
- Feature Map Extraction with 5 different resolution levels
- Image Embedding Generation
- Transfer Learning Support
Frequently Asked Questions
Q: What makes this model unique?
This model combines EfficientNet's compound scaling with enhanced training procedures (RA2), offering superior efficiency-accuracy balance compared to traditional architectures. It's particularly notable for its optimized training recipe that includes advanced augmentation and optimization strategies.
Q: What are the recommended use cases?
The model excels in image classification tasks, feature extraction for downstream tasks, and as a backbone for transfer learning. It's particularly suitable for applications requiring a good balance between computational efficiency and accuracy, with support for both high-resolution inference and memory-constrained scenarios.
